multi-objective optimization of building envelope

ASHRAE and IBPSA-USA SimBuild 2016Building Performance Modeling Conference

Salt Lake City, UTAugust 8-12, 2016

Multi-Objective Optimization of Building Envelope, Lighting and HVAC Systems Designs

Weili Xu1, Khee Poh Lam1, Adrian Chong1, Omer T. Karaguzel11Center for Building Performance and Diagnostics, Carnegie Mellon University, Pittsburgh, PA

ABSTRACTThis paper proposes a computational building design sup-port framework. The framework evaluates and optimizesbuilding envelope systems, electrical systems and HVACsystems simultaneously with the assistance of a newly de-veloped EnergyPlus component cost database. The inte-grated building system evaluation could potentially revealthe trade-offs among different building systems and helpclients and design teams a gain better understanding ofhigh performance designs.A case study based on a real office building is presentedto demonstrate the effectiveness of this framework. Ener-gyPlus v8.3 was used as the building performance evalu-ation tool. The NSGA-II multi-objective optimization al-gorithm was employed for acquiring the best design com-binations. Results obtained from the optimization processproved that buildings with high performance componentswill present savings for both first and operation costs.Trade-offs between the first cost and operation costs fordifferent designs were also discussed within five selectedoptimized cases.

INTRODUCTIONThe building industry is notoriously driven by the firstcost. Life cycle cost analysis (LCCA) might change thedesign decision-making by forcing the consideration ofthe net present value of long-term operation costs How-ever, this type of analysis does not usually reveal the trade-offs between short-term investment and long-term sav-ings, which are important economic details for buildingstakeholders. To further enhance the process of designdecision making, multi-objective building system designoptimization, that minimizes first and operation costs si-multaneously at the whole building level, could be an ef-fective approach.However, several challenges exist for this approach. Oneof them is to develop a first cost estimation methodology.Currently, two approaches are frequently employed in re-search and industry projects. They are the comparativecost estimation method and the quantity take-off (QTO)method. Comparative cost estimation evaluates a project’scosts based on its type, size, location, etc. Machine learn-ing techniques are employed in this method as cost pre-

dictors. However, such an approach is commonly usedat the early design stage when only a conceptual designis available (Gunnaydin and Dogan 2004). Unlike com-parative cost estimation, QTO can be adopted at most ofthe design stages. This cost estimation technique focuseson measuring a building’s schematics or work results,which is collected in the bill of quantities (Monteiro andMartins 2013). However, traditional QTO requires man-ual measurement building components from 2D drawings,which is labor intensive and error-prone. With the growthof Building Information Modeling (BIM) techniques, 3Dbased BIM-QTO is becoming a preferred process, whichoffers quick and accurate project cost estimation (Eastmanet al. 2011). Despite its popularity, this process is alsofacing huge challenges. A survey conducted in 2011 indi-cate that more than 80% of BIM models received by gen-eral contractors do not contain adequate data for perform-ing high quality cost estimation (Sattineni and Bradford2011). The omission of important information will cer-tainly jeopardize the BIM-QTO automation process andforce quantity surveyors to move back to the manual pro-cess. In addition, it is impossible to calculate building op-eration costs in this process without integrating buildingenergy simulations.

Recent studies propose methods that couple multi-objective optimization with building energy simulation toevaluate project economic impacts (Evins 2013). Thesemethods utilize the computational power of building en-ergy models to estimate both building energy performanceprofiles and long-term economic impacts. However, someof these studies focus on optimizing a specific system’sperformance with regard to a single or multiple objectives.Under such conditions, the optimized system could likelybecome sub-optimal when integrated with other systemsdue to the lack of a synergistic relationship between dif-ferent building systems. On the other hand, studies fol-lowing a holistic approach for optimizing design param-eters usually fix the HVAC system type as a constant in-put or a set of workarounds is implemented to incorpo-rate HVAC impact on the model response. By fixing theHVAC system type, numerical optimization can only takeinto account some typical component performance param-eters such as chiller COPs or boiler thermal efficiencies.

© 2016 ASHRAE (www.ashrae.org). For personal use only. Additional reproduction, distribution, or transmission in either print or digital form is not permitted without ASHRAE's prior written permission.

110

Such an isolated optimization approach could possibly re-sult in unrealistic product parameters that a designer cannever find in the actual product lines of the market. Inaddition, workaround approaches can lead to inaccurateperformance evaluations.This paper presents a method to optimize both first andoperation costs of building designs simultaneously withrespect to building envelope, electrical lighting and HVACsystem design options. A newly developed HVAC auto-generation engine is incorporated in this method to varyHVAC system types during optimization runtime. In ad-dition, the new engine can automatically size system ca-pacities based on energy model settings and design daydata. With these two features, the method provides real-istic cost estimations for the integrated building systemsas well as maintenance and replacement schedules. Byconsidering the interactions between building envelope,lighting and HVAC systems, the whole building simula-tion can reveal relationships among different design cate-gories thus providing designers a better understanding oftheir design proposals and return on investment. Ener-gyPlus is used as the energy simulation engine. A JAVA-based application is introduced to link the building systemdata, generate EnergyPlus HVAC systems and to performthe NSGA-II optimization algorithm with the use of theJMetal package (Durillo and Nebro 2011).

METHODFigure 2 shows an overview of the design support frame-work. The framework consists of five main mod-els namely Building Energy Model (BEM), EnergyPlusHVAC auto generation module, Construction Classifica-tion Model (CCM), Cost Optimization Model and BEM-QTO Mapping Model. The BEM-QTO Mapping Mod-ule is developed to map BEM and CCM for buildingcomponent cost selection. The additional inputs to thisframework are design options, which contain user selecteddesign alternatives for the building, and life cycle costmodel. The life cycle cost model carries economic evalua-tion parameters such as interest rate, price escalation rate,tax rate etc.

Building Energy ModelIn the proposed optimization framework, BEM is basedon EnergyPlus V8.3 and contains not only building con-struction and building electric systems, but also HVACsystems. EnergyPlus is one of the advanced whole build-ing simulation engines. It includes comprehensive build-ing components’ descriptions regarding building enve-lope, electrical and HVAC systems. In addition, its lifecycle cost model is well-developed and tested by the Pa-cific Northwest National Laboratory (PNNL) based on theNational Institute of Standards and Technology (NIST)Handbook 135 standard (Cho et al. 2011). Therefore,

EnergyPlus allows building owners and designers holisti-cally appraise different design alternatives for their energyand operation costs.In this framework, a building energy model in EnergyPlusprovides building system information to the BEM-QTOMapping Module for cost mapping. It also computes thewhole building life cycle cost once the cost estimationprocess is completed.

HVAC Auto-generation Model

Due to the complexity of HVAC system modeling in Ener-gyPlus, a HVAC auto-generation engine is achieved by anewly developed thermal zone name convention, a HVACXML data schema and an idf translator that translatesXML data into .idf format. The zone name conventionand HVAC XML data schema will be detailed in the fol-lowing sections.

Zone Name

In a building energy model, thermal zone names shouldbe designed to carry both architectural and HVAC sys-tem information. A new naming convention is proposedthat allows tools to identify each thermal zone prop-erly. Five elements are required in a zone’s name (Fig-ure 1). They are block, zone function, zone identifica-tion, zone’s mechanical ventilation group and its ther-mal condition group. The first three elements store azone’s architectural information and the last two ele-ments indicate a zone’s HVAC system information. Forinstance, Floor6%Office%601%DOAS6%VRF6E2 indi-cates the number 601 thermal zone is located on the 6th

floor and it is an office room. Regarding to its mechanicalsystem information, this zone belongs to the DOAS6 zonegroup for mechanical ventilation and the VRF6E2 zonegroup for indoor thermal condition. There are some re-served characters for the last two elements, which are de-signed for the zones that require special mechanical con-ditions. For example, an ”EXT” in the ventilation groupelement means an exhaust system is required in this zone.This zone name convention allows tools to quickly con-

Figure 1: Zone Naming Convention

struct basic zone activity assumptions and some commonHVAC systems. The current tested HVAC systems rangefrom packaged terminal air conditioner, variable air vol-ume (VAV) systems, to hybrid dedicated outdoor air sys-tem (DOAS) and variable refrigerant flow (VRF) systemsetc.


111

Figure 2: BEM-QTO Design Decision Support Framework

HVAC Data StructureThe HVAC data is stored in an XML format. The designof this XML data schema focuses on providing quick dataconversion between EnergyPlus .idf data structures andHVAC performance data structures. The design quality at-tributes of the HVAC system XML data schema includes:

1. Adaptable to common HVAC systems’ performancedata

2. Extensible and easy to modify with EnergyPlus ver-sion upgrades.

The XML format has elements include:<dataset>: dataset is the parent element that containsthe entire data of one specific HVAC system. It has twoattributes:

1. setname: specify the HVAC product name of thisdataset

2. version: specify EnergyPlus version for this dataset

Example: <dataset setname = “Manufactuer VRF 4Ton”version = “V8.3”/>.<object>: object is a child of <object>element. Thiselement represents the objects in EnergyPlus. It has twoattributes:

1. description: the name of the EnergyPlus object

2. reference: the object reference. This attribute indi-cates the sub-system that this object belongs to.

Example: <object description = “AirLoopHVAC” refer-ence = “Supply Side System”>.<field>: field is a child element of <object>. This el-ement represents the individual field under an object inEnergyPlus. It also has two attributes:

1. description: the key of a field

2. type: data type, this includes “String”, “Double” and“Integer”.

Example: <field description = “Gross Rated CoolingCOP” type = “Double”>3.2</field>.The data schema can directly link to the EnergyPlusHVAC system model based on a two-layer logic pro-cess. The first layer identifies HVAC system type andEnergyPlus version to extract the corresponding datasetat dataset element. The next step applies auto genera-tion strategies to different sub-systems. For instance, thesupply side system is created by analyzing both the venti-lation group, and the thermal condition group in the zonename.


112

Construction Classification ModelIn Figure 2, the structure of CCM is programmed basedon the MasterFormat 2004 classification standard. Mas-terFormat is one of the leading trade-based industry clas-sification standards developed for construction work thatare specified jointly by the Construction Specifications In-stitute (CSI) and the Construction Specifications Canada(CSC). The classification standard has a 50-year history inthe industry, and gained the most recognition for organiz-ing commercial construction specifications in the NorthAmerican region (CSI and CSC 2005). More impor-tantly, several construction cost data providers such asRSMeans R© adopt this standard for constructing their costdatabase (Gulledge et al. 2007) (Keady 2013). Therefore,in this study, the cost data structure is developed accordingto MasterFormat 2004 classification and cost data is ex-tracted from RSMeans Building Construction Cost Data(RSMeans 2015).The developed cost model firstly accepts building compo-nent specifications from the BEM-QTO Mapping Mod-ule, and performs the cost mapping to its database. Themapped cost vector (including material, labor, equipment,total) is then sent back to the BEM-QTO Mapping processfor cost estimation.

BEM-QTO Mapping ModuleThe BEM-QTO Mapping process is developed for au-tomating the first cost estimation. This process is made ofa BEM-QTO mapping system, which implements a datacommunication protocol to unify both CCM and BEMmodels’ semantic, and a products cost selection module,to achieve two major objectives respectively: 1. Ensuringa seamless data mapping flow among models; 2. Ensuringthe selected building products and their costs are real andavailable.The mapping system consists of three building system li-braries: building envelope system, building HVAC systemand building electrical system. The libraries are designedto include EnergyPlus data schema as well as cost map-ping elements required for first cost estimation. In addi-tion, each building component can be linked to its ownmaintenance, repair and replacement (MRR) items in thedatabase. The MRR data will then be exported to Energy-Plus LifeCycleCost:RecurringCosts object for life cyclecost analysis. More information about this object can befound in the EnergyPlus Input/Output reference (LBNL2013).

OptimizationThe framework focuses on design decision support fromthe economic perspective of the project. Therefore, firstand operation costs are identified as the study objectivesand evaluated in a multi-objective optimization process.Among the popular multi-objective optimization algo-

rithms, the NSGA-II is selected in this study. This al-gorithm is a well-recognized nondominated sorting basedalgorithm. It is designed to improve its predecessor’s per-formance speed while keeping the diversity of the opti-mal solutions set. It also shows the ability to convergenear to the true pareto-optimal front (Deb et al. 2002).The algorithm has been tested on nine different problems.The results demonstrated that the NSGA-II is more effec-tive than the pareto-archived evolution strategy (PAES)and the strength-pareto evolutionary algorithm (SPEA)on most of these problems. Due to its superior per-formance, the NSGA-II is collected in the JMetal pack-age, a Java based multi-objective optimization framework(Durillo and Nebro 2011). By accessing the JMetal’s API,the BEM-QTO framework could utilize the power of theNSGA-II algorithm to evaluate different design optionsand form a set of optimal solutions.In the BEM-QTO framework, the building system designproblem can be formulated as a multi-objective optimiza-tion problem with discrete type parameters:

min{ f1(x), f2(x)} (1)

Where f1(x) is the operation costs, which is stated:

f1(x) = xrep− xres + xe + xom&r (2)

f2(x) is the first cost and it can be formulated as:

f2(x) = xci + xhi + xei + xrest (3)

Where xi (first cost), xci (envelope costs), xhi (HVAC sys-tem costs), xei (electrical system costs), xrest (other costs),xrep (present value of capital replacement costs), xres(present value of residual value less the disposal costs),xe (present value of energy costs), xom&r (present value ofnon-fuel operating, maintenance and repair costs). Equa-tion 2 and 3 are adapted from NIST Handbook 135 withno water cost included (Fuller and Petersen 1995).Theconstraint of this optimization process is:

X ={

x ∈ Zsc|xi j, i ∈ {1..c} , j ∈ {1..s}}

(4)

The Zsc is the constraint set, which contains the pre-defined building system types (s) and their cost informa-tion (c). The design parameter (x) is specified as discreteindependent variables that can only take the values fromZsc.

CASE STUDYThe building selected for this study is an office buildinglocated in Pittsburgh, Pennsylvania. It consists of fourfloors, a penthouse floor and a basement with a total of3700 m2 floor area. The building’s energy model is builtin DesignBuilder v4.5 and exported to EnergyPlus v8.3.Figure 3 below shows the geometry of this building asrepresented in EnergyPlus .dxf file.


113

Figure 3: Representation of Building in EnergyPlus

Table 1: NSGA-II Parameter ValuesParameter ValueCrossover SinglePointCrossOver:

90% probabilityMutation BitFlipMutation: 17%

probabilitySelection BinaryTournament2

(Deb et al. 2002)Population 30Maximum Evaluation 900

Design Variables

The designed mechanical conditioning in this building issupplied via 3 air handling units (AHU) and four-pipe fancoil units located within each room. AHU 1 is a VAVsystem that supplies conditioned air to the basement andfirst floor through terminal units with reheat coil. AHU 2is a VAV system that supplies conditioned air only to theauditorium. AHU 3 supplies ventilation to the 2nd, 3rdand 4th floors. Heating and cooling for 2nd, 3rd and 4thfloors are provided via a four-pipe fan coil system. Be-sides the designed VAV system, VRF, and hybrid DOASand VRF system are also introduced as possible designoptions for this building. In the design evaluation, theother two HVAC systems’ configurations are kept iden-tical to the designed VAV system.

Table 2: Life Cycle Cost Model ParametersField InputDiscounting Convention End of YearInflation Approach Constant DollarReal Discount Rate 0.03Base Date 2015 JanuaryService Date 2017 JanuaryStudy Length 25 YearsTax Rate 0.06Depreciation Method StraightLine-27 YearsPrice Escalation NIST Handbook 135 Ta-

ble Ca-1, Census Region1

Electricity Tariff Duquesne Light Rate GSNatural Gas Tariff Equitable Gas Rate GSS

Besides HVAC systems, there are five other design op-tions, namely external wall and roof constructions, win-dow systems, lighting fixtures and daylighting control op-tions. External wall and roof constructions are built ac-cording to Table 16 and 17 in ASHRAE Handbook 2009Chapter 18 (ASHRAE 2009). Six window system designsare created in WINDOW 6.3 (Table 3). These design op-tions vary from low performance double layer window tohigh performance quadruple window. Lighting design op-tions contain not only the typical T5, T8 fluorescent light-ing fixtures, but also LEDs. Furthermore, occupancy con-trol is also introduced in the evaluation (Table 3). In to-tal, there are 71,820 possible design combinations in thisstudy. A desktop with configuration of i7 quad-core 3.5GHz CPU and 16 GB RAM is used in this study. EachEnergyPlus simulation takes about 8 minutes. A brute-force search approach would require more than 400 daysto complete a full evaluations.

Study Assumptions

Compare to brute-force search approach, the optimizationalgorithm implemented in the framework can greatly re-duce the total evaluation time down to a reasonable level.The set up of the NSGA-II can be found in Table 1. Theprobability of crossover operation is set to the JMetalpackage, NSGA-II default value. Mutation probability iscalculated by 1/N where N is the number of design vari-ables. Each generation evaluates 30 design solutions andthe maximum number of evaluation is limited to 900. Intotal, there are 30 generations in the optimization process.Parallel processing strategy is applied in this framework,which allows computer to evaluate 6 energy simulationsconcurrently. This strategy could greatly reduce the timerequired for finding the best design solutions. The totalevaluation time takes about 23 hours.Table 2 indicates the parameters used in the life cycle cost


114

Table 3: Window and Lighting Design OptionDesign Property Unit CostWindow: Dou-ble Clear (DC)

U-3.13, SHGC-0.73, Vt-0.8

$242.19/m2

Window: Dou-ble Tinted (DT)

U-2.58, SHGC-0.37, Vt-0.53

$290.36/m2

Window: Dou-ble Thick Clear(DTC)

U-1.4, SHGC-0.41, Vt-0.61

$356.82/m2

Window: HeatReflective Clear(HRC)

U-1.4, SHGC-0.25, Vt-0.45

$478.99/m2

Window: TripleGlazing (TG)

U-0.81, SHGC-0.71, Vt-0.53

$480.00/m2

Window:Quadruple (Q)

U-0.781, SHGC-0.46, Vt-0.62

$862.00/m2

Light: T8 10.2W/m2 $149.5/m2

Light: T5 8W/m2 $163.5/m2

Light: LED 6.5W/m2 $389.5/m2

OccupancySensor

$268/Each

Note: The unit of U value is W/m2 ◦C

model. These parameters are defined according to the an-nual supplement report of NIST Handbook 135 (Rushing,Kneifel, and Lavappa 2015). As part of the life cycle costcomponent, a utility cost model is built based on the lo-cal utility provider’s tariff. Unlike average energy cost($/kWh), the utility cost model can effectively capture theimpact of energy demand in different designs. In Ener-gyPlus, it can accurately identify the cost tariff based onbuilding peak demand for placing excessive energy de-mand charges ($/kW). This implies that not only a build-ing energy consumption, but also its peak demand, couldaffect the building’s operation costs.

Results and Discussion

Figure 4 shows the design solutions’ distribution from theoptimization process. The red dots indicate solutions gen-erated at initial generation and purple dots show the solu-tions generated at last generation. The X-axis is the oper-ations costs of the project and the Y-axis is the first costof the project. Both of the costs are valued in USD ($).At initial stage, a wide dispersion of red dots suggest thatthe the NSGA-II algorithm are exploring the entire solu-tion space. As the process continues, the design cases areslowly converging to the region at the lower left cornerof the solution space, which shows lower first and opera-tion costs. The adequacy of 900 evaluations is also tested.Figure 5 compares a few selected pareto front curves gen-erated along the optimization process with the final paretofront curve. At the beginning, the pareto front curve at 2nd

Figure 4: Distribution of design solutions generated bythe NSGA-II process

●●●

● ●

●

●●

●

●

● ●●●

● ●

●●●

●

●●

●●●●●

●

●●

●

●

●●

●

●

●●●●●

●● ●

●●● ●●

●

●

●●●

●

● ●

●

●

●

●●

●●

1250000

1500000

1750000

2000000

2250000

2500000

1800000 2000000 2200000Operation Cost ($)

Firs

t Cos

t ($)

Generation

●

●

2

30

●●

●

●

●●●●●

●

● ●

●●● ●●

●

●

●●●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●●●●●

●

● ●

●●● ●●

●

●

●●●

●

● ●

●

●

●

●●

●

●

1250000

1500000

1750000

2000000

1800000 1900000 2000000 2100000 2200000Operation Cost ($)

Firs

t Cos

t ($)

Generation

●

●

29

30

Figure 5: Pareto front curves comparison

generation shows a large deviation from the final paretofront curve. As the generation increases, the generatedpareto front curves become closer to the final pareto frontcurve. The lower graph shows a small deviation betweenthe pareto front curve at generation 29 and the final paretofront curve, which shows that the final pareto front curvecould be close to the global pareto front curve.Besides solutions generated from optimization, a basecase model is created that represents the initial design.The initial design uses design options that have the lowestnormalized unit cost. In the final pareto front curve, 30design options are suggested. Among them, 6 cases areselected for comparison. They are the highest first cost(Case 26), lowest first cost (Case 13), highest operationcosts (Case 25), lowest operation cost (Case 4), highestlife cycle cost (Case 26) and lowest life cycle cost (Case1). Table 4 shows the design solutions for each of the se-lected case and their correspondent first cost as well as op-eration costs. In Table 4, both base case’s first cost and op-eration costs are higher than the optimized cases’. For firstcost, though the base case model installs all the cheapestdesign options, its HVAC sizing results suggest a largeamount of initial investment on the mechanical system. In


115

Table 4: Optimal Solution Packages

Case Walla Roofa Window Light Daylight HVAC First Cost($)

OperationCosts ($)

ABEFC($)

Base 24 9 DC T 8 Off VAV 2,286,433 2,324,763 01 9 3 T G LED On VRF 1,309,218 1,945,480 1,356,4984 34 17 T G LED On DVb 2,114,405 1,808,939 687,85213 7 7 HRC LED Off VRF 1,260,953 2,160,938 1,189,30525 22 16 T G T 5 On VRF 1,287,545 2,256,787 1,066,86426 9 4 T G LED On DVb 2,181,651 1,810,763 618,782

a: Wall and Roof number is based on ASHRAE Handbook 2009 Fundamental Chapter 18 Table 16 and 17.b: DV denotes hybrid DOAS and VRF system.

addition, due to its lower performance design, base case’sEUI is nearly 60% higher than optimized cases, whichlead to higher operation costs.Since all the optimized cases have lower first and opera-tion costs, there is no payback period for the project in-vestment. Therefore, Additional Break-Even First Cost(ABEFC) analysis is used in this study. ABEFC is afirst cost based analysis, which converts all the savings inthe project period to first cost. With this method, projectstakeholders can compare optimized cases by evaluatinghow much additional first cost could be justified to breakeven the base case (Mumma 2002). In this study, theABEFC is equal to the life cycle cost difference betweenoptimized cases and base case. In Table 4, Case 1 isnot only the lowest life cycle cost case, but also has themost additional first cost. With almost $1.4 million saved,clients or design teams could invest on high efficiencyequipment or design elements that can create a more com-fortable working environment. Besides Case 1, other opti-mized cases also achieve a large amount of additional firstcost. Besides the additional first cost, clients and designteams can also compare the trade-offs between first andoperation costs among the optimized cases by visualizingthe solutions on pareto front curve. Figure 6 shows a pos-sible decision making scenario. The upper graph indicatesthis project’s first and operation costs can be reduced by42% and 16% respectively by moving from base case (reddot) to case 25. Trade-offs between optimized cases canalso be visualized on the same graph. The lower graphin Figure 6 shows a 4% increase in first cost and 13% de-crease in operation costs by changing from case 25 to case1. By visualizing the results, clients and design teams canadjust their design choice based on project budget and pre-dicted operation period.Table 4 also suggests the most common systems for op-timized cases. They are triple glazing, LED, occupancylight control and VRF systems. These systems are theexpensive design options; however, they can effectivelyreduce building energy consumption and building peakdemand. A smaller peak demand means a smaller and

Figure 6: Decision making process

less complicated HVAC system. If the savings from theHVAC system is significant, the incremental costs fromenvelope and electrical systems can be minimized, or evenreduced. Besides the first cost savings, integrating highperformance envelope, electrical and HVAC systems canpotentially yield a lower life cycle cost as well.

CONCLUSIONA BEM-QTO framework for building systems design de-cision making is proposed in this study. The following listsummarizes the features of this framework:

1. Project budget can be estimated according to thenational construction classification using the BEMmodel

2. With the newly developed HVAC data schema, dif-ferent HVAC systems can be automatically generatedin EnergyPlus according to the predefined ventilationgroups and thermal condition groups.

3. The NSGA-II algorithm is demonstrated to be ef-fective in building system design optimization. The


116

algorithm can find nearly optimal design solutionswhile greatly reducing the evaluation time.

4. Building life cycle cost analysis can be performedat a integrative level, which not only considers theperformance of each individual system, but also theinteractions among the building systems.

5. The framework offers design solution packages withconsideration of the first cost as well as operationcosts for each solution.

6. The generated design solutions are realistic, whichcan be directly drawn from the results and used fordetail cost estimation and the bill of quantities.

The current framework is limited to building systems andcomponent evaluation. Therefore, building system con-trol strategies are not considered in this study. This is notonly because the cost of system control strategies is dif-ficult to estimate, but also the evaluation requires mixedinteger and double data type, which could fundamentallychange the behaviors of operators in the NSGA-II algo-rithm. Further development of this implementation is un-der progress. In addition, because building energy simu-lation require extensive computational power, the numberof building system design options should be limited to arelatively small set to avoid a large amount of evaluationtime. Therefore, expert knowledge is required before per-forming this method. Lastly, several critical cost compo-nents such as ducts and pipes are not considered in the firstcost estimation process. This is because BEM does notprovide such information in the simulation results. Po-tential future work could connect this framework with afully specified Building Information Model, which offersquantity counts for such building elements.

REFERENCESASHRAE. 2009. ASHRAE Handbook 2009 Fundamen-

tal. Atlanta, GA: American Society of Heating, Re-frigerating and Air-Conditioning Engineers.

Cho, H, W Wang, A Makhmalbaf, K Yun, J Glazer,L Scheier, V Srivastava, and K Gowri. 2011. “Ex-tend EnergyPlus to Support Evaluation, Design, andOperation of Low Energy Buildings.” Technical Re-port, Pacific Northwest National Laboratory.

CSI, and CSC. 2005. “MasterFormat TM 2004 EditionNumbers & Tiles.” Technical Report, ConstructionSpecifications Institute and Construction Specifica-tions Canada.

Deb, Kalyanmoy, Amrit Pratap, Sameer Agarwal, andTAMT Meyarivan. 2002. “A fast and elitist multiob-jective genetic algorithm: NSGA-II.” EvolutionaryComputation, IEEE Transactions on 6 (2): 182–197.

Durillo, Juan J., and Antonio J. Nebro. 2011. “jMetal:A Java framework for multi-objective optimization.”Advances in Engineering Software 42:760–771.

Eastman, Chuck, paul Teicholz, Sacks Rafael, and Kath-leen Liston. 2011. A BIM Application Areas ForOwners. Hoboken, New Jersey: John Wiley & SonsInc.

Evins, Ralph. 2013. “A review of computational op-timisation methods applied to sustainable buildingdesign.” Renewable and Sustainable Energy Review22:230–245.

Fuller, S K, and S R Petersen. 1995. NIST Handbook135: Life-Cycle Costing Manual for the Federal En-ergy Management Program. Gaithersburgh, MD:National Institute of Standards and Technology.

Gulledge, Charles E, Lane J Beougher, Michael J King,Robert Paul Dean, Dennis J Hall, Nina M Giglio,G Wade Bevier, and Phil Steinberg. 2007. “Master-Format 2004 Edition 2007 Implementation Assess-ment.” Technical Report, Continental AutomatedBuilding Association.

Gunnaydin, H Murat, and S Zeynep Dogan. 2004. “Aneural network approach for early cost estimation ofstructural systems of buildings.” International Jour-nal of Project Management, p. 595.

Keady, R A. 2013. Equipment Inventories for Ownersand Facility Managers: Standards, Strategies andBest Practices. Hoboken, New Jersey: John Wiley& Sons Inc.

LBNL. 2013. “Input Output Reference: The Encyclo-pedic Reference to EnergyPlus Input and Output.”Technical Report, Lawrence Berkeley National Lab-oratory.

Monteiro, Andre, and Joao Pocas Martins. 2013. “Asurvey on modeling guidelines for quantity takeoff-oriented BIM-based design.” Automation in Con-struction, pp. 238–253.

Mumma, Stanley A. 2002. “Using Dedicated OutdoorAir Systems: Economics of improved environmentalquality.” ASHRAE Journal 1:1–10.

RSMeans. 2015. Building Construction Cost Data. Nor-well, MA: Construction Publishers & Consultants.

Rushing, Amy S, Joshua D Kneifel, and Priya Lavappa.2015. “Energy Price Indices and Discount Factorsfor Life Cycle Cost Analysis 2015.” Technical Re-port, National Institute of Standards and Technology.

Sattineni, Anoop, and R Harrison Bradford. 2011. “Es-timating with BIM: A survey of US ConstructionCompanies.” ISARC Conference, Seoul, Korea.


117

multi-objective optimization of building envelope

Documents